1

The Navier–Stokes Equations

Remark 1.1. Basic principles and variables. The basic equations of fluid dy-
namics are called Navier–Stokes equations. In the case of an isothermal flow, a
flow at constant temperature, they represent two physical conservation laws –
the conservation of mass and the conservation of linear momentum. There are
various ways for deriving these equations. Here, the classical one of continuum
mechanics will be used. Let the flow variables be
• ρ(t, x) : density [kg/m3 ],
• v(t, x) : velocity [m/s],
• P (t, x) : pressure [N/m2 ],
which are assumed to be sufficiently smooth functions in the time interval
[0, T ] and the domain Ω ⊂ R3 . 2

Fig. 1.1. Left: Claude Louis Marie Henri Navier (1785 – 1836), right: George Gabriel
Stokes (1819 – 1903).
4 1 The Navier–Stokes Equations

1.1 The Conservation of Mass

Remark 1.2. General conservation law. Let V be an arbitrary open volume
in Ω with sufficiently smooth surface ∂V which is constant in time and with
mass Z
m(t) = ρ(t, x) dx, [kg].
V
If mass in V is conserved, the rate of change of mass in V must be equal to
the flux of mass ρv(t, x) [kg/(m2 s)] across the boundary ∂V of V

d d
Z Z
m(t) = ρ(t, x) dx = − (ρv) (t, s) · n(s) ds, (1.1)
dt dt V ∂V

where n(s) is the outward pointing unit normal on s ∈ ∂V . Since all functions
are assumed to be sufficiently smooth, the divergence theorem can be applied
(integration by parts), which gives
Z Z
∇ · (ρv) (t, x) dx = (ρv) (t, s) · n(s) ds.
V ∂V

Inserting this identity into (1.1) leads to

Z  
∂ρ
(t, x) + ∇ · (ρv) (t, x) dx = 0.
V ∂t

Since V is an arbitrary volume, it follows

 
∂ρ
+ ∇ · (ρv) (t, x) = 0 ∀ t ∈ (0, T ], x ∈ Ω. (1.2)
∂t

This is the first equation of mathematical fluid dynamics, which is called

continuity equation. 2

Remark 1.3. Incompressible, homogeneous fluid. If the fluid is incompressible

and homogeneous, i.e., composed of one fluid only, then ρ(t, x) = ρ0 > 0
and (1.2) reduces to

(∂x v1 + ∂y v2 + ∂z v3 ) (t, x) = ∇ · v(t, x) = 0 ∀ t ∈ (0, T ], x ∈ Ω, (1.3)

where  
v1 (t, x)
v(t, x) =  v2 (t, x)  .
v3 (t, x)
Thus, the conservation of mass for an incompressible, homogeneous fluid im-
poses a constraint on the velocity only. 2
 1.2 The Conservation of Linear Momentum 5

1.2 The Conservation of Linear Momentum

Remark 1.4. Newton’s second law of motion. The conservation of linear mo-
mentum is the formulation of Newton’s second law of motion
net force = mass × acceleration
for flows. It states that the rate of change of the linear momentum must be
equal to the net force acting on a collection of fluid particles. 2

Fig. 1.2. Left: Isaac Newton (1642 – 1727), right: Brook Taylor (1685 – 1731).

Remark 1.5. Acceleration. Consider a fluid particle at time t and position x

with velocity v(t, x) and a small time interval ∆t. A linear extrapolation
of the particle path gives that the particle has at time t + ∆t the position
x + ∆tv, see Figure 1.3. The acceleration of the particle is, using again a
linear approximation (first order Taylor series expansion in time),
dv v (t + ∆t, x + ∆tv(t, x)) − v(t, x)
(t, x) = lim
dt ∆t→0 ∆t
v(t, x) + ∆tvt (t, x) + ∆t (v(t, x) · ∇) v(t, x) − v(t, x)
≈ lim
∆t→0 ∆t
∂v
= (t, x) + (v(t, x) · ∇) v(t, x)
∂t
= vt (t, x) + (v(t, x) · ∇) v(t, x).
In the linear Taylor series approximation with respect of the second ar-
gument, the increment, which is needed in the Taylor expansion, is ∆tv(t, x)
and the derivative has to be taken with respect to x for each component of
the velocity
   
v(t, x) · ∇v1 (t, x) v1 ∂x v1 + v2 ∂y v1 + v3 ∂z v1
∆t  v(t, x) · ∇v2 (t, x)  = ∆t  v1 ∂x v2 + v2 ∂y v2 + v3 ∂z v2  , (1.4)
v(t, x) · ∇v3 (t, x) v1 ∂x v3 + v2 ∂y v3 + v3 ∂z v3
6 1 The Navier–Stokes Equations

where the explicit dependency on t and x has been neglected in the right term
of (1.4) for clarity of presentation. In the usual notation (v · ∇)v, one thinks
of v · ∇ = v1 ∂x + v2 ∂y + v3 ∂z acting on each component of v. This expression
is the same as (1.4) without ∆t. In the literature, one often finds the notation
v · ∇v. The gradient of the velocity is a tensor with the components
∂vi
(∇v)ij = ∂j vi = , i, j = 1, 2, 3.
∂xj

Altogether, the first order approximation is used as model for ’mass ×

acceleration’ in an arbitrary volume V
Z
ρ(t, x) (vt + (v · ∇)v) (t, x) dx [N ].
V

This expression must be balanced by the net forces acting on V which are
composed of external (body) forces and internal forces. 2

Remark 1.6. External forces. External forces include, e.g., gravity, buoyancy
and electromagnetic forces (in liquid metals). These forces are collected in a
body force term Z
F(t, x) dx, F : [N/m3 ].
V
2

Remark 1.7. Internal forces, Cauchy’s principle and the stress tensor. Internal
forces are forces which a fluid exerts on itself in trying to get out of its own
way. These include pressure and viscous drag that a fluid element exerts on
the adjacent element. The internal forces of a fluid are contact forces, i.e., they
act on the surface of the fluid element V . Let t [N/m2 ] denote this internal
force vector, which is called Cauchy stress vector or torsion vector, then the
contribution of the internal forces on V is

Fig. 1.3. Movement of a fluid particle.

 1.2 The Conservation of Linear Momentum 7
Z
t(t, s) ds.
∂V

Thus, the equation for the conservation of linear momentum is, for an arbitrary
constant-in-time volume V ,
Z Z Z
ρ(t, x) (vt + (v · ∇)v) (t, x) dx = F(t, x) dx + t(t, s) ds. (1.5)
V V ∂V

The right hand side of (1.5) describes the net force acting on and inside V .
Now, a detailed description of the internal forces represented by t(t, s) is
necessary.
The foundation of continuum mechanics is the stress principle of Cauchy.
The idea of Cauchy on internal contact forces was that on any (imaginary)
plane on ∂V there is a force that depends (geometrically) only on the orien-
tation of the plane. Thus, t = t(n), where n is the unit normal vector of the
imaginary plane which points outward of V . It can be shown that, if linear
momentum is conserved, t is a linear function of n, i.e.,

t = Sn, (1.6)

where S(t, x) [N/m2 ] is a 3 × 3-matrix (tensor) which is called stress tensor.

The stress tensor represents all internal forces of the flow. Inserting (1.6) into
the term representing the internal forces in (1.5) and applying the divergence
theorem give Z Z
t(t, s) ds = ∇ · S(t, x) dx,
∂V V

where the divergence of a matrix (tensor) is defined row-wise

(a11 )x + (a12 )y + (a13 )z

   
∂x a11 + ∂y a12 + ∂z a13
∇ · A =  (a21 )x + (a22 )y + (a23 )z  =  ∂x a21 + ∂y a22 + ∂z a23  .
(a31 )x + (a32 )y + (a33 )z ∂x a31 + ∂y a32 + ∂z a33

Since (1.5) holds for every volume V , it follows

ρ (vt + (v · ∇)v) = ∇ · S + F ∀ t ∈ (0, T ], x ∈ Ω. (1.7)

This is the momentum equation. 2

Remark 1.8. Symmetry of the stress tensor. Let V be an arbitrary volume with
sufficiently smooth boundary ∂V and let the net force given by the right hand
side of (1.5). The torque in V with respect to the origin 0 of the coordinate
system is then defined by
Z Z
M0 = r × F dx + r × (Sn) ds [N m]. (1.8)
V ∂V
8 1 The Navier–Stokes Equations

Fig. 1.4. Left: Augustin Louis Cauchy (1789 – 1857), right: Robert Hooke (1635 –
1703).

Here, r = x1 e1 + x2 e2 + x3 e3 is the vector pointing from 0 to a point x ∈ V .

A straightforward calculation shows

r × (Sn) = (r × S∗1 r × S∗2 r × S∗3 ) n,

where S∗i is the i-th column of S and (·) denotes here the matrix. Inserting
this expression into (1.8) and applying integration by parts leads to
Z Z
M0 = r × F dx + ∇ · ((r × S∗1 r × S∗2 r × S∗3 )) dx (1.9)
ZV V
Z
= r × (F + ∇ · S) dx + ∂x r × S∗1 + ∂y r × S∗2 + ∂z r × S∗3 dx.
V V

Consider now a fluid in equilibrium state, i.e. the net forces acting on this
fluid are zero. Hence, the right hand side of (1.7) vanishes and so the first
integral of (1.9). In addition, equilibrium requires in particular that M0 = 0.
Thus, from (1.9) follows
Z
0= ∂x r × S∗1 + ∂y r × S∗2 + ∂z r × S∗3 dx. (1.10)
V

Using now
(x1 + ∆x1 )e1 − x1 e1
∂x r = lim = e1 ,
∆x1 →0 ∆x1
∂y r = e2 , ∂z r = e3 , and inserting these equations into (1.10) leads finally to
 
Z S32 − S23
0=  S13 − S31  (t, x) dx.
V S21 − S12

for an arbitrary volume V . It follows that S has to be symmetric, S = ST , and

S possesses six unknown components. 2
 1.2 The Conservation of Linear Momentum 9

Remark 1.9. Decomposition of the stress tensor. To model the stress tensor in
the basis variables introduced in Remark 1.1, the stress tensor is decomposed
into
S = V + P I.
Here, V [N/m2 ] is the so-called viscous stress tensor, representing the forces
coming from the friction of the particles, and P [N/m2 ] is the pressure, de-
scribing the forces acting on the surface of each fluid volume V . The viscous
stress tensor will be modeled in terms of the velocity, see Remark 1.11. 2

Remark 1.10. The pressure. The pressure P acts on a surface only normal to
that surface of a fluid volume and it is directed into the volume V . Thus, the
total force exerted by the pressure which is acting on the volume is
Z Z Z
− P n ds = − ∇P dx = − ∇ · (P I) dx.
∂V V V

Remark 1.11. The viscous stress tensor. Friction between fluid particles can
only occur if the particles move with different velocities. For this reason, the
viscous stress tensor depends on the gradient of the velocity. For the reason
of symmetry, Remark 1.8, it depends on the symmetric part of the gradient,
the so-called velocity deformation tensor
T
∇v + (∇v)
D (v) = .
2
If the velocity gradients are not too large, one can assume that the dependency
is linear. This leads to the model
 
2µ
V = 2µD (v) + ζ − (∇ · v) I, (1.11)
3

where µ [kg/(m s)] and ζ [kg/(m s)] are the first and second order viscosities
of the fluid. The viscosity µ is also called dynamic or shear viscosity. The
law (1.11) is the analog for fluids of Hooke’s law for solids. 2

Remark 1.12. Newtonian fluids. The linear relation (1.11) is only an approx-
imation for a real fluid. In general, the relation will be non-linear. Only for
small stresses, a linear approximation of the general stress-deformation rela-
tion can be used. The first scientist to postulate a linear stress-deformation
relation was Newton. For this reason, a fluid satisfying assumption (1.11) is
called Newtonian fluid. More general relations than (1.11) exist, however they
are less well understood. 2

Remark 1.13. Normal and shear stresses, trace of the stress tensor. The diag-
onal components of the stress tensor S11 , S22 , S33 are called normal stresses
and the off-diagonal components shear stresses .
10 1 The Navier–Stokes Equations

The trace of the stress tensor is the sum of the normal stresses

tr(S) = S11 + S22 + S33

 
2µ
= µ(∂x v1 + ∂y v2 + ∂z v3 ) + 3 ζ − (∇ · v) + 3P
3
= (3ζ − µ) ∇ · v + 3P.

For incompressible flows, it follows

1
P (t, x) = (S11 + S22 + S33 ) (t, x).
3
2

Remark 1.14. The Navier–Stokes equations with dimensions. Now, the pres-
sure part of the stress tensor and the ansatz (1.11) of the viscous stress tensor
can be inserted into (1.7) giving the general Navier–Stokes equations (includ-
ing the conservation of mass)

ρ (vt + (v · ∇)v)
2µ



−2∇ · (µD (v)) − ∇ · ζ− 3 ∇ · vI + ∇P = F in (0, T ] × Ω,
ρt + ∇ · (ρv) = 0 in (0, T ] × Ω.

If the fluid is incompressible and homogeneous, such that µ is a constant, the

Navier–Stokes equations simplify to
P F
vt − 2ν∇ · D (v) + (v · ∇)v + ∇ = in (0, T ] × Ω,
ρ0 ρ0 (1.12)
∇ · v = 0 in (0, T ] × Ω.

Here, ν = µ/ρ0 [m2 /s] is the kinematic viscosity of the fluid. 2

1.3 The Dimensionless Navier–Stokes Equations

Remark 1.15. Characteristic scales. Mathematical analysis and numerical sim-
ulations are based on dimensionless equations. The functions in system (1.12)
are not dimensionless. To derive a dimensionless equations, the quantities
• L [m] – a characteristic length scale of the flow problem,
• U [m/s] – a characteristic velocity scale of the flow problem,
• T ∗ [s] – a characteristic time scale of the flow problem,
are introduced. Let (t0 , x0 ) [s, m] be the old variables. 2

Remark 1.16. The Navier–Stokes equations in dimensionless form. Applying

the transform of variables
x0 v t0
x= , u= , t= ,
L U T∗
 1.3 The Dimensionless Navier–Stokes Equations 11

one obtains from (1.12) and a rescaling

L 2ν P L
ut − ∇ · D (u) + (u · ∇)u + ∇ = F in (0, T ] × Ω,
UT∗ UL ρ0 U 2 ρ0 U 2
∇·u = 0 in (0, T ] × Ω,
where all derivatives are with respect to the new variables. Defining
P UL L L
p= , Re = , St = , f= F, (1.13)
ρ0 U 2 ν UT∗ ρ0 U 2
the Navier–Stokes equations in dimensionless form
2
Stut − ∇ · D (u) + (u · ∇)u + ∇p = f in (0, T ] × Ω,
Re (1.14)
∇ · u = 0 in (0, T ] × Ω
are obtained. The constant Re is called Reynolds number and the constant
St Strouhal number. These numbers allow the classification and comparison
of different flows. 2

Fig. 1.5. Left: Osborne Reynolds (1842 – 1912), right: Čeněk Strouhal (1850 –
1923).

Remark 1.17. Simplified notation of the dimensionless Navier–Stokes equa-

tions. To simplify the notations, one uses the characteristic quantities L =
1 m, U = 1 m/s, and T ∗ = L/U = 1 s such that (1.14) simplifies to
ut − 2ν∇ · D (u) + (u · ∇)u + ∇p = f in (0, T ] × Ω,
(1.15)
∇ · u = 0 in (0, T ] × Ω,

with the dimensionless viscosity ν = Re−1 .

This transform and the resulting equations (1.15) will be the basis equa-
tions for the mathematical analysis and the numerical simulation of the
Navier–Stokes equations. There are two important difficulties for the mathe-
matical analysis and the numerical simulation of the Navier–Stokes equations:
12 1 The Navier–Stokes Equations

• the coupling of velocity and pressure,

• the non-linearity of the convective term.
Additionally, difficulties for the numerical simulation occur if
• the convective term dominates the viscous term, i.e. ν is small.
2
Remark 1.18. Different forms of terms in (1.15). With the help of the di-
vergence constraint, i.e. the second equation in (1.15), the viscous and the
convective term of the Navier–Stokes equations can be reformulated equiva-
lently.
Assume, u is sufficiently smooth with ∇ · u = 0. Then
 
(∇ · u)x
∇ · (∇u) = ∆u, ∇ · ∇uT =  (∇ · u)y  = 0.

(∇ · u)z

Thus, the viscous term becomes

−2ν∇ · D (u) = −ν∆u.

For the convective term, ones uses the identity (product rule)

∇ · uvT = (∇ · v)u + (v · ∇)u.

In the case v = u with ∇ · u = 0, it follows

(u · ∇) u = ∇ · uuT .

Note that different forms of the terms are in general not longer equivalent
for a discretization of the Navier–Stokes equations since the discrete velocity
field is in general not divergence-free. 2
Remark 1.19. Two-dimensional Navier–Stokes equations. Even if real flows
occur only in three dimensions, the consideration of the Navier–Stokes equa-
tions (1.15) in two dimensions is also of interest. E.g., there are applications
where the flow is constant in the third direction and it behaves virtually two-
dimensional. 2
Remark 1.20. Special cases of incompressible flow models.
• In a stationary flow, the velocity and the pressure do not change in time.
Hence ut = 0 and these flows are modelled by the so-called stationary or
steady-state Navier–Stokes equations
−ν∆u + (u · ∇)u + ∇p = f in Ω,
(1.16)
∇ · u = 0 in Ω.

A necessary condition for the stationarity of a flow field is that the data
of the problem, i.e. the right hand side and the boundary conditions, see
Sect. 1.4, are time-independent. But this condition is not sufficient.
 1.4 Initial and Boundary Conditions 13

• If in a stationary flow the viscous transport dominates the convective trans-

port, i.e. if the fluid flows very slowly, the non-linear convective term of
the Navier–Stokes equations (1.16) can be neglected. This gives a linear
equation, the so-called Stokes equations

−∆u + ∇p = f in Ω,
(1.17)
∇ · u = 0 in Ω.

Here, the momentum equation was divided by ν, defining a new pressure

and a new right hand side.
• In the numerical analysis, often the so-called Oseen equations are con-
sidered. Given a divergence-free flow field u0 , the Oseen equations are a
system of linear equations of the form

−ν∆u + (u0 · ∇)u + ∇p + cu = f in Ω,

(1.18)
∇ · u = 0 in Ω,

with a real number c ≥ 0.

2

Fig. 1.6. Left: Carl Wilhelm Oseen (1879 – 1944), right: Johann Peter Gustav
Lejeune Dirichlet (1805 – 1859).

1.4 Initial and Boundary Conditions

Remark 1.21. General. The Navier–Stokes equations (1.15) are a first order
partial differential equation with respect to time and a second order partial dif-
ferential equation with respect to space. Thus, they have to be equipped with
an initial condition at t = 0 and with boundary conditions on the boundary
Γ = ∂Ω of Ω, if Ω is a bounded domain. There are several kinds of bound-
ary conditions which can be prescribed for incompressible flows. Of course, a
14 1 The Navier–Stokes Equations

compatibility condition should be fulfilled between the boundary conditions

of the initial velocity field and the limit of the prescribed boundary conditions
for t → 0, t > 0. 2

Remark 1.22. Initial condition. Concerning the initial condition, an initial ve-
locity field, which has to be divergence-free, is prescribed at t = 0

u(0, x) = u0 (x) with ∇ · u0 = 0 in Ω.

Remark 1.23. Dirichlet boundary conditions, no-slip boundary conditions. An

often used boundary conditions describes the velocity field on a part of the
boundary
u(t, x) = g(t, x) in (0, T ] × Γdiri ⊂ Γ.
This boundary condition is called Dirichlet boundary condition. It models
in particular prescribed inflows into Ω and outflows from Ω. In the special
case g(t, x) = 0 in (0, T ] × Γdiri , this boundary condition is called no-slip
boundary condition. Let n be the unit normal vector in x ∈ Γnosl ⊂ Γdiri
and {t1 , t2 } unit tangential vectors such that {n, t1 , t2 } is an orthonormal
system of vectors. Then, the no-slip boundary condition can be decomposed
into three parts:

u(t, x) = 0 ⇐⇒ u(t, x) · n = 0, u(t, x) · t1 = 0, u(t, x) · t2 = 0

in x ∈ Γnosl . The condition u(t, x) · n = 0 states that the fluid does not
penetrate the wall. The other two conditions describe that the fluid does not
slip along the wall.
If Dirichlet boundary conditions are prescribed on the whole boundary
of Ω, the pressure is determined only up to an additive constant. An additional
condition for fixing the constant has to be introduced, e.g., that the integral
mean value of the pressure should vanish
Z
p(t, x) dx = 0 t ∈ (0, T ].
Ω

Remark 1.24. Free slip boundary conditions, slip with friction boundary con-
ditions. The free slip boundary condition is applied on boundaries without
friction. It has the form

u·n = g in (0, T ] × Γslip ⊂ Γ,

T
n Stk = 0 in (0, T ] × Γslip , 1 ≤ k ≤ d − 1. (1.19)

There is no penetration through the wall if g = 0 on Γslip .

 1.4 Initial and Boundary Conditions 15

Fig. 1.7. James Clerk Maxwell (1831 – 1879)

The slip with linear friction and no penetration boundary condition has
the form

u·n = 0 in (0, T ] × Γslfr ⊂ Γ,

u · tk + β −1 nT Stk = 0 in (0, T ] × Γslfr , 1 ≤ k ≤ d − 1. (1.20)

This boundary condition states that the fluid does not penetrate the wall and
it slips along the wall whereas it loses energy. The loss of energy is given by
the friction parameter β. In the limit case β −1 → 0, the no-slip condition is
recovered and in the limit case β −1 → ∞ the free slip condition. Slip with
friction boundary conditions were studied already by Maxwell [Max79] and
Navier [Nav23]. The difficulty in the application of this boundary condition
consists in the determination of the friction parameter β, which might depend,
e.g., on the local flow field and on the roughness of the wall.
Since n and tk are orthogonal vectors, the values of the pressure do not play
any role in the boundary conditions (1.19) and (1.20). Hence, an additional
condition for the pressure is needed to fix the additive constant. 2

Remark 1.25. Outflow or do-nothing boundary conditions. For numerical sim-

ulations, the so-called outflow boundary condition or do-nothing boundary
condition
Sn = 0 in (0, T ] × Γoutf ⊂ Γ
is often applied. This boundary condition models that the normal stress, which
is equal to the Cauchy stress vector (1.6), vanishes on the boundary part Γoutf .
The do-nothing boundary condition is used in flow problems, e.g., where no
other outflow boundary condition is available. 2

Remark 1.26. Conditions for an infinite domain, periodic boundary condi-

tions. The case Ω = R3 is also considered in analytical and numerical studies
of the Navier–Stokes equations. There are two situations in this case. In the
first one, the decay of the velocity field as kxk2 → ∞ is prescribed. The
second situation consists of periodic boundary conditions. These boundary
16 1 The Navier–Stokes Equations

conditions do not posses any physical meaning. They are used to simulate an
infinite extension of Ω in one or more directions. Let, e.g., this direction be
ei . It is assumed that the flow is periodic in this direction with the length
l of the period. In computations, e.g., the cube Ω = (0, l)d is used and the
periodic boundary conditions are given by

u(t, x + lei ) = u(t, x) ∀ (t, x) ∈ (0, T ] × Γ.

From the point of view of the finite computational domain, all appearing
functions have to be extended periodically in the periodic direction to return
to the original problem.
The use of space periodic boundary conditions may also facilitate analyt-
ical investigations, see Temam [Tem95, p. 4]. 2